CT image processing method and apparatus for detecting and segmenting perforator in CT image

Abstract

Disclosed are a CT image processing method and a CT image processing apparatus for detecting and segmenting a perforator in a CT image. The CT image processing method may comprise the steps of: detecting, from a CT image, a region of interest (ROI) expected to have a perforator, by using an ROI detection model; generating a plurality of three-dimensional patches including at least a portion of the detected ROI; and inputting the plurality of three-dimensional patches to a perforator detection model and generating a result image obtained by extracting the perforator from the CT image on the basis of a segmentation result from the perforator detection model.

Description

CT image processing method and device for detecting and segmenting perforations in CT images

[0001] The following disclosure relates to a CT image processing technique for detecting and segmenting perforations in CT images.

[0002] Before performing skin reconstructive surgery, it is crucial to identify the location of the blood supply (perforator) to the entire flap. However, traditionally, this process has been performed manually by medical professionals. The accuracy of this manual assessment, which relies solely on the surgeon's skill level, is limited. Therefore, inexperienced medical professionals can reduce the accuracy of blood supply location assessment, and failure to accurately identify blood supply locations can lower the success rate of skin reconstructive surgery. Therefore, research is needed to overcome these limitations.

[0003] A CT image processing method for detecting and segmenting perforations in a CT (computed tomography) image according to one embodiment may include the steps of: detecting a region of interest (ROI) in which a perforation is expected to exist from the CT image using a region of interest detection model; generating a plurality of three-dimensional patches including at least a portion of the detected region of interest; and inputting the plurality of three-dimensional patches into a perforation detection model and generating a result image by extracting the perforation from the CT image based on a segmentation result from the perforation detection model.

[0004] The above region of interest detection model and the perforation detection model may include a neural network model, and the step of detecting the region of interest may include a step of detecting the region of interest for each frame of the CT image; and a step of determining a region of interest in a three-dimensional shape in the shooting area of ​​the CT image based on the region of interest detected for each frame.

[0005] The step of generating the three-dimensional patches may include a step of generating the three-dimensional patches by dividing the three-dimensional region of interest into three-dimensional sub-regions smaller than the region of interest.

[0006] The above-mentioned perforation detection model can generate a classification result that classifies the generated 3D patches into a perforation region, which is an area including the perforation, and a general region, which is an area not including the perforation, and output the segmentation result that segments the perforation region from the CT image based on the classification result.

[0007] The step of generating the above result image may include a step of generating the result image in which the perforation site is indicated in the CT image by masking the CT image based on the segmentation result.

[0008] The above-mentioned perforation detection model includes a softmax layer and a convolution layer, and the softmax layer generates a classification result that classifies a perforation area including the perforation area and a general area not including the perforation area from the generated 3D patches, and the convolution layer can output a segmentation result that segments the perforation area based on the classification result.

[0009] According to one embodiment, a CT image processing device for detecting and segmenting a perforation site in a computed tomography (CT) image may include: a region of interest detection unit that detects a region of interest (ROI) in which a perforation site is expected to exist from the CT image using a region of interest detection model; a three-dimensional patch generation unit that generates a plurality of three-dimensional patches including at least a portion of the detected region of interest; and a result image generation unit that inputs the plurality of three-dimensional patches to a perforation site detection model and generates a result image by extracting the perforation site from the CT image based on a segmentation result from the perforation site detection model.

[0010] The above region of interest detection model and the perforation detection model include a neural network model, and the region of interest detection unit can detect the region of interest for each frame of the CT image and determine a region of interest in a three-dimensional shape in the shooting area of ​​the CT image based on the region of interest detected for each frame.

[0011] The above 3D patch generation unit can generate the 3D patches by dividing the 3D region of interest into 3D sub-regions smaller than the 3D region of interest.

[0012] The above-mentioned perforation detection model can generate a classification result that classifies the generated 3D patches into a perforation region, which is an area including the perforation, and a general region, which is an area not including the perforation, and output the segmentation result that segments the perforation region from the CT image based on the classification result.

[0013] The above result image generation unit can generate the result image in which the perforation site is indicated in the CT image by masking the CT image based on the segmentation result.

[0014] The above-mentioned perforation detection model includes a softmax layer and a convolution layer, and the softmax layer generates a classification result that classifies a perforation area including the perforation area and a general area not including the perforation area from the generated 3D patches, and the convolution layer can output a result that segments the perforation area based on the classification result.

[0015] According to one embodiment, the accuracy of identifying the location of a blood vessel or perforation site, which was previously done manually, can be improved by automating the process.

[0016] In one embodiment, the success rate of skin reconstructive surgery can be improved by accurately identifying the location of blood vessels or perforations.

[0017] FIG. 1 is a diagram illustrating an overview of a CT image processing system according to one embodiment.

[0018] Figure 2 is a flowchart for explaining a CT image processing method according to one embodiment.

[0019] FIG. 3 is a drawing for explaining a CT image processing method according to one embodiment.

[0020] FIG. 4 is a diagram illustrating a configuration of a CT image processing device according to one embodiment.

[0021] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Therefore, the actual implementation is not limited to the specific embodiments disclosed, and the scope of this specification includes modifications, equivalents, or alternatives within the technical concepts described in the embodiments.

[0022] Although terms such as "first" or "second" may be used to describe various components, these terms should be interpreted solely to distinguish one component from another. For example, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component.

[0023] When it is said that a component is "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but there may also be other components in between.

[0024] Singular expressions include plural expressions unless the context clearly dictates otherwise. In this document, phrases such as "A or B", "at least one of A and B", "at least one of A or B", "A, B, or C", "at least one of A, B, and C", and "at least one of A, B, or C" can each include any one of the items listed together in that phrase, or all possible combinations thereof. In this specification, it should be understood that the terms "comprises" or "has" and the like are intended to specify the presence of a described feature, number, step, operation, component, part, or combination thereof, but do not exclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0025] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art. Terms defined in commonly used dictionaries should be interpreted to have a meaning consistent with their meaning in the context of the relevant technology, and will not be interpreted in an idealized or overly formal sense unless explicitly defined herein.

[0026] The term "module" as used herein may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit. A module may be an integral component, or a minimum unit or portion of such a component that performs one or more functions. For example, according to one embodiment, a module may be implemented in the form of an application-specific integrated circuit (ASIC).

[0027] The term "~part" as used in this document refers to a software or hardware component such as an FPGA or ASIC, and the "~part" performs certain roles. However, the "~part" is not limited to software or hardware. The "~part" may be configured to reside on an addressable storage medium and may be configured to execute one or more processors. For example, the "~part" may include components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within the components and "~parts" may be combined into a smaller number of components and "~parts" or further separated into additional components and "~parts." Furthermore, the components and "~parts" may be implemented to execute one or more CPUs within a device or a secure multimedia card. Additionally, '~bu' may include one or more processors.

[0028] Hereinafter, embodiments will be described in detail with reference to the attached drawings. In the description with reference to the attached drawings, identical components are assigned the same reference numerals regardless of the drawing numbers, and redundant descriptions thereof will be omitted.

[0029] FIG. 1 is a diagram illustrating an overview of a CT image processing system according to one embodiment.

[0030] The CT image processing system described herein relates to a diagnostic assistance technology for artificial intelligence medical image data, and more specifically, to a deep learning algorithm that can detect and segment perforators of a flap for skin reconstruction using the anterolateral thigh in medical CT images or dental CT images. The CT image processing system can provide a deep learning algorithm technology that can automatically detect and segment the blood flow of perforators of an anterolateral thigh free flap with high accuracy. It is important to accurately detect the perforators of a patient's anterolateral thigh flap and the course of the perforators for skin reconstruction using the anterolateral thigh, in which the free flap of the anterolateral thigh is transplanted to another site. More specifically, it is important to accurately detect the perforators and the course of the perforators of a flap of the anterolateral thigh of a patient in order to properly connect the blood vessels of the flap taken from the anterolateral thigh with the blood vessels of the site to be transplanted. The CT image processing system described herein can provide a CT image processing method that can automatically detect perforations of an anterolateral femoral free flap from a patient's CT image with high accuracy using deep learning technology. Although the CT image processing system and the CT image processing method are described herein based on an embodiment for the anterolateral thigh, the CT image processing system and the CT image processing method are not limited to the embodiments described herein.

[0031] Referring to FIG. 1, a CT image processing system may include a CT image processing device (110) that performs a CT image processing method for detecting and segmenting perforations in a CT (computed tomography) image.

[0032] A CT image processing device (110) can receive a CT image (120). The CT image processing device (110) can detect a region of interest, in which a perforation is expected to exist, from the CT image (120) using a region of interest detection model. The CT image processing device (110) can generate a plurality of three-dimensional patches including at least a portion of the detected region of interest. The CT image processing device (110) can input the plurality of three-dimensional patches into the perforation detection model. In response to inputting the plurality of three-dimensional patches into the perforation detection model, the CT image processing device (110) can obtain a classification result and a segmentation result from the perforation detection model. The classification result may be a classification of the generated three-dimensional patches into a perforation region, which is an region including a perforation, and a general region, which is an region not including a perforation, and the segmentation result may be a segmentation of the perforation region from the CT image (120) based on the classification result. The CT image processing device (110) can generate a result image (130) by extracting a perforation in a CT image (120) based on at least one of the classification result and the segmentation result.

[0033] Below, a CT image processing method performed by a CT image processing device can be described in more detail using drawings.

[0034] The perforation described in this specification may also be referred to as perforation blood circulation, blood circulation, blood circulation of perforation, blood vessel, perforation blood vessel, blood vessel of perforation, and is not limited to the description in this specification.

[0035] Figure 2 is a flowchart for explaining a CT image processing method according to one embodiment.

[0036] Referring to FIG. 2, in step (210), a CT image processing device may detect a region of interest (ROI) in which a perforation is expected to exist from a CT image using a region of interest detection model. For example, the CT image processing device may detect a region of interest in which a perforation is expected to exist in the form of a margin box, a bounding box, or a square box using the region of interest detection model. For example, the CT image processing device may detect a region of interest such that the upper boundary of the region of interest is located at the anterior superior iliac spine (ASIS) and the lower boundary of the region of interest is located at the superior border of the patella. The CT image processing device may detect a region of interest for each frame of the CT image. The CT image processing device may determine a region of interest in a three-dimensional shape in the shooting area of ​​the CT image based on the region of interest detected for each frame. Here, each frame of the CT image may be two-dimensional, and each frame of the CT image may be stacked to form a three-dimensional CT image. Here, the region of interest detection model may include a neural network model. A region of interest detection model can be trained using deep learning to improve the accuracy of identifying regions of interest, including perforated areas. The region of interest detection model can focus deep learning-based detection only within specific regions, enabling high accuracy and efficiency in region designation. Examples of region of interest detection models include, but are not limited to, YOLOv7.

[0037] In step (220), the CT image processing device can generate a plurality of three-dimensional patches including at least a portion of the detected region of interest. The CT image processing device can generate the three-dimensional patches by dividing the three-dimensional region of interest into three-dimensional sub-regions smaller than the region of interest. The CT image processing device can generate or obtain the plurality of three-dimensional patches by dividing the three-dimensional region of interest into a plurality of three-dimensional sub-regions smaller than the region of interest. The three-dimensional patches can be generated or obtained in the shape of a hexahedron, for example, but the shape of the three-dimensional patches is not limited to that described in the present specification.

[0038] In step (230), the CT image processing device can input a plurality of three-dimensional patches into a perforation detection model, and generate a result image in which perforations are extracted from the CT image based on the segmentation result from the perforation detection model. The perforation detection model can include a neural network model, and for example, can include ResNet, but is not limited thereto. The perforation detection model can generate a classification result in which the generated three-dimensional patches are classified into a perforation region, which is an area including a perforation, and a general region, which is an area not including a perforation. The perforation detection model can output a segmentation result in which the perforation region is segmented from the CT image based on the classification result. The CT image processing device can generate a result image in which perforations are displayed in the CT image by masking the CT image based on the output segmentation result. In other words, the CT image processing device can generate a result image in which perforations are displayed in the CT image by masking the CT image based on the segmented perforation region included in the segmentation result. The resulting image may appear as if only the perforation site is highlighted in the CT image, with the rest of the image masked. The resulting image may also be referred to as a masked image.

[0039] The perforation detection model may include a softmax layer and a convolutional layer. The softmax layer may generate or output a classification result that classifies a perforation region including a perforation region and a general region not including a perforation region from the generated 3D patches. The convolutional layer may output a segmentation result that segments the perforation region based on the classification result. The perforation detection model may perform classification and segmentation simultaneously. In addition, the perforation detection model may be trained in a multitask-learning manner that simultaneously learns classification and segmentation during training. The perforation detection model may be trained to accurately classify a perforation region, which is an area including a perforation region, and a general region, which is an area not including a perforation region. In addition, the perforation detection model may be trained to accurately segment the perforation region from a CT image based on the classification result.

[0040] FIG. 3 is a drawing for explaining a CT image processing method according to one embodiment.

[0041] Referring to FIG. 3, in step (310), a CT image processing device can detect a region of interest (ROI) (311, 312, 313) in which a perforation is expected to exist from a CT image (314, 315) using a region of interest detection model (316).

[0042] A CT image processing device can detect a region of interest (311, 312) from each frame (314) of a CT image. The CT image processing device can determine a region of interest in a three-dimensional shape in a shooting area of ​​the CT image based on the region of interest detected for each frame including the frame (314). The three-dimensional region of interest can be detected, such as a region of interest (313), in a cross-section (315) in a direction perpendicular to the frame (314) of the CT image in a three-dimensional CT image in which each frame of the CT image is stacked.

[0043] The region of interest detection model (316) can detect or designate a region of interest by detecting the location of the perforation site. The region of interest detection model (316) can be trained to detect a region of interest by detecting the location of the perforation site. The region of interest detection model (316) can include YOLOv7 and can be trained using a segmentation mask in which the perforation site of the anterolateral femoral free flap is segmented by an expert as training data or label data. During the training process, the region of interest detection model (316) can receive training data or label data and identify the perforation site based on the input training data or label data. The region of interest detection model (316) can designate or detect the region of interest based on the location of the perforation site identified based on the training data or label data as a center point. The region of interest can be referred to as a bounding box or a margin box. The region of interest designated during training of the region of interest detection model (316) may also be referred to as a second region of interest to distinguish it from the region of interest designated by the region of interest detection model (316) during the process of performing the CT image processing method. The region of interest designated during the process of performing the CT image processing method may be referred to as a first region of interest. The region of interest may have a horizontal length and a vertical length. Accuracy may be calculated based on the region of interest designated by the region of interest detection model (316). More specifically, classification information regarding whether a perforation site is included within the region of interest or within the range of a location where the region of interest is designated may be calculated. Accuracy for the region of interest designated by the region of interest detection model (316) may be measured based on the classification information. Here, the accuracy may correspond to the accuracy of how accurately the perforation site is included in the region of interest designated by the region of interest detection model (316).The accuracy of a region of interest can be measured using the Intersection over Union (IoU) metric. Classification information can be calculated using the cross-entropy loss function. Accuracy measured using the IoU metric can be an indicator of the degree of overlap between two regions. In other words, accuracy measured using the IoU metric can be an indicator of the degree of overlap between the perforated paper and the region of interest.

[0044] In step (320), the CT image processing device can generate a plurality of three-dimensional patches including at least a portion of the detected region of interest (311, 312, 313). The CT image processing device can generate the three-dimensional patches by dividing the detected region of interest (311, 312, 313) into three-dimensional sub-regions smaller than the region of interest (311, 312, 313).

[0045] The CT image processing device can classify and segment the detailed range of the perforation site in the region of interest (311 312, 313) based on a three-dimensional patch. In step (330), the CT image processing device can input a plurality of three-dimensional patches to the perforation site detection model (337) and generate a result image (334) by extracting the perforation site (332, 333) from within the CT image (331) based on the segmentation result from the perforation site detection model (337). The CT image processing device can also input a plurality of three-dimensional patches to the perforation site detection model (337) and generate a result image (334) by extracting the perforation site (332, 333) from within the CT image (331) based on at least one of the classification result and the segmentation result from the perforation site detection model (337).

[0046] The perforation detection model (337) may include a ResNet, and the last layer of the ResNet may include two different output layers. These two different output layers may be a softmax layer (339) and a convolutional layer (338). The softmax layer (339) may output a classification result, and the convolutional layer (338) may generate a segmentation result.

[0047] The hole detection model (337), more specifically, the softmax layer (339) of the hole detection model (337), can generate a classification result that classifies a hole area including a hole and a general area not including a hole from 3D patches. The hole detection model (337), or the softmax layer (339) of the hole detection model (337), can output or generate a classification result that classifies a hole area including a hole and a general area not including a hole by classifying whether or not a hole is present in each 3D patch while windowing in units of 3D patches.

[0048] The perforation detection model (337), more specifically, the convolution layer (338) of the perforation detection model (337), can output a segmentation result that segments the perforation area from the CT image based on the classification result. That is, the perforation detection model (337) or the convolution layer (338) of the perforation detection model (337) can output a segmentation result that segments the perforation area from the CT image based on the perforation area and the general area. The segmentation described in this specification may also be referred to as, for example, segmentation, division, or words similar to the meaning of the English word 'segmentation', and is not limited to the examples in this specification. The perforation detection model (337) may perform classification and segmentation at the same time, and learning of classification and learning of segmentation may also be performed at the same time. The perforation detection model (337) can segment how the range or shape of the perforation is drawn. The perforation detection model (337) may include, for example, a ResNet model, and may perform segmentation on the detected perforation. The perforation detection model (337) may be trained to accurately perform segmentation on the detected perforation. The perforation detection model (337) may receive 3D patches in which the region of interest is segmented into a 3D shape as input, and may be trained using a segmentation mask in which the perforation of the anterolateral femoral free flap is segmented by an expert as training data or label data, similar to the region of interest detection model (316). The perforation detection model (337) may output or generate a classification result that classifies a perforation region including a perforation region and a general region not including a perforation region from the 3D patches, and may be trained to measure and improve the accuracy of the output classification result based on the training data or label data. The accuracy of the classification result generated or output by the perforation detection model (337) may be measured on a pixel basis using Dice loss.Accuracy measured using die loss can be an indicator of how much overlap there is between two regions. That is, accuracy measured using die loss can be an indicator of how much overlap there is between the segmentation result and the perforation field.

[0049] At this time, in order to improve the accuracy, an evaluation is performed on the classification result generated or output by the perforation detection model (337), and parameters for generating or outputting the classification result by the perforation detection model (337) can be adjusted based on the evaluation. Parameters for the perforation detection model (337) can be adjusted so that it can generate or output a classification result that accurately classifies the general area and the perforation area. In addition, the perforation detection model (337) can output a segmentation result that segments the perforation area from the CT image based on the classification result, and the output segmentation result can be trained so that the accuracy can be measured and the accuracy can be improved based on learning data or label data. At this time, in order to improve the accuracy, an evaluation is performed on the segmentation result output by the perforation detection model (337), and parameters for outputting the segmentation result by the perforation detection model (337) can be adjusted based on the evaluation. Parameters for the perforation detection model (337) can be adjusted so that it can output a segmentation result that accurately segments the perforation area from the CT image.

[0050] The classification result output or generated by the perforation detection model (337) during the learning process may be a different classification result from the classification result generated or output by the perforation detection model (337) during the execution of the CT image processing method. Therefore, the classification result output or generated by the perforation detection model (337) during the learning process may be referred to as a second classification result, and in this case, the classification result generated or output by the perforation detection model (337) during the execution of the CT image processing method may be referred to as a first classification result. Similarly, the segmentation result output by the perforation detection model (337) during the learning process may be a different segmentation result from the segmentation result output by the perforation detection model (337) during the execution of the CT image processing method. Therefore, the segmentation result output by the perforation detection model (337) during the learning process may be referred to as a second segmentation result, and in this case, the segmentation result output by the perforation detection model (337) during the execution of the CT image processing method may be referred to as a first segmentation result.

[0051] The CT image processing device can generate a result image (334) based on the segmentation result. The CT image processing device can generate a result image (334) in which the perforations (335, 336) are displayed in the CT image by masking the CT image based on the segmented perforation area included in the segmentation result. The CT image processing device can generate a result image (334) in which only the perforations (335, 336) are displayed as if they are highlighted in the CT image and other areas except for the perforations (335, 336) are masked and covered. The perforations (335, 336) in the result image (334) can be displayed as perforations (332, 333) in the CT image (331), more specifically, in each frame (331) of the CT image.

[0052] FIG. 4 is a diagram illustrating a configuration of a CT image processing device according to one embodiment.

[0053] Referring to FIG. 4, the CT image processing device (400) may include a region of interest detection unit (410), a 3D patch generation unit (420), and a result image generation unit (430). The operations of the region of interest detection unit (410), the 3D patch generation unit (420), and the result image generation unit (430) described in the present disclosure may be performed by a device including a processor and a memory. The CT image processing device (400) may correspond to the CT image processing device described in the present specification. In one embodiment, the operations of the region of interest detection unit (410), the 3D patch generation unit (420), and the result image generation unit (430) may be performed by one or more processors. The one or more processors may be, for example, a central processing unit (CPU), a graphic processor unit (GPU), a neural processor unit (NPU), or any combination thereof. One or more processors may be arranged and operated on an integrated circuit (IC) or an application-specific integrated circuit (ASIC).

[0054] The region of interest detection unit (410) can detect a region of interest (ROI) in which a perforation is expected to exist from a CT image using a region of interest detection model. The region of interest detection unit (410) can detect a region of interest for each frame of the CT image, and determine a three-dimensional region of interest in the shooting area of ​​the CT image based on the region of interest detected for each frame. The region of interest detection unit (410) can detect a region of interest such that, for example, the upper boundary of the region of interest is located at the anterior superior iliac spine (ASIS) and the lower boundary of the region of interest is located at the superior border of the patella. The region of interest detection unit (410) can detect a region of interest for each frame of the CT image.

[0055] The 3D patch generation unit (420) can generate a plurality of 3D patches including at least a portion of the detected region of interest. The 3D patch generation unit (420) can generate the 3D patches by dividing the 3D region of interest into 3D sub-regions smaller than the region of interest. The 3D patch generation unit (420) can generate or obtain a plurality of 3D patches by dividing the 3D region of interest into 3D sub-regions smaller than the region of interest.

[0056] The result image generation unit (430) can input a plurality of 3D patches into a perforation detection model, and generate a result image by extracting perforations from a CT image based on the segmentation result from the perforation detection model. The perforation detection model can generate a classification result that classifies the generated 3D patches into a perforation region, which is an area including a perforation, and a general region, which is an area not including a perforation. The perforation detection model can output a segmentation result that segments the perforation region from the CT image based on the classification result. The result image generation unit (430) can generate a result image in which perforations are displayed in the CT image by masking the CT image based on the segmentation result. The result image generation unit (430) can generate a result image in which perforations are displayed in the CT image by masking the CT image based on the segmented perforation region included in the segmentation result. The result image can be displayed as if only perforations are emphasized in the CT image, and other areas except for the perforations are masked.

[0057] The embodiments described above may be implemented using hardware components, software components, and / or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may be implemented using a general-purpose computer or a special-purpose computer, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing instructions and responding to them. The processing device may execute an operating system (OS) and software applications running on the operating system. Furthermore, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing device is sometimes described as being used alone; however, one of ordinary skill in the art will recognize that the processing device may include multiple processing elements and / or multiple types of processing elements. For example, a processing unit may include multiple processors, or a processor and a controller. Other processing configurations, such as parallel processors, are also possible.

[0058] Software may include computer programs, codes, instructions, or a combination of one or more of these, which may configure a processing device to perform a desired operation or may independently or collectively command the processing device. The software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium, or device for interpretation by the processing device or for providing instructions or data to the processing device. The software may also be distributed over networked computer systems and stored or executed in a distributed manner. The software and data may be stored on a computer-readable recording medium.

[0059] The method according to the embodiment may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may store program commands, data files, data structures, etc., alone or in combination, and the program commands recorded on the medium may be those specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program commands, such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

[0060] The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

[0061] The term "module" as used herein may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit. A module may be an integral component, or a minimum unit or portion of such a component that performs one or more functions. For example, according to one embodiment, a module may be implemented in the form of an application-specific integrated circuit (ASIC).

[0062] The term "~part" as used in this document refers to a software or hardware component such as an FPGA or ASIC, and the "~part" performs certain roles. However, the "~part" is not limited to software or hardware. The "~part" may be configured to reside on an addressable storage medium and may be configured to execute one or more processors. For example, the "~part" may include components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within the components and "~parts" may be combined into a smaller number of components and "~parts" or further separated into additional components and "~parts." Furthermore, the components and "~parts" may be implemented to execute one or more CPUs within a device or a secure multimedia card. Additionally, '~bu' may include one or more processors.

[0063] Although the embodiments described above have been described with limited drawings, those skilled in the art will appreciate that various technical modifications and variations can be applied based on the described embodiments. For example, appropriate results can still be achieved even if the described techniques are performed in a different order than described, and / or components of the described systems, structures, devices, circuits, etc. are combined or combined in a different manner than described, or are replaced or substituted with other components or equivalents.

[0064] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims described below.

Claims

1. A CT image processing method for detecting and segmenting perforations in CT (computed tomography) images, A step of detecting a region of interest (ROI) in which a perforation is expected to exist from the CT image using a region of interest detection model; a step of generating a plurality of three-dimensional patches including at least a portion of the detected region of interest; and A step of inputting the plurality of three-dimensional patches into a perforation detection model and generating an image by extracting the perforation from the CT image based on the segmentation result from the perforation detection model. A CT image processing method comprising:

2. In paragraph 1, The above region of interest detection model and the above perforation detection model are, Contains a neural network model, The step of detecting the above region of interest is: A step of detecting the region of interest for each frame of the CT image; and A step of determining a three-dimensional region of interest in the shooting area of ​​the CT image based on the region of interest detected for each frame above. A CT image processing method comprising:

3. In paragraph 2, The step of generating the above 3D patches is: A step of generating the three-dimensional patches by dividing the three-dimensional region of interest into three-dimensional sub-regions smaller than the three-dimensional region of interest. A CT image processing method comprising:

4. In paragraph 1, The above perforation detection model is, Generating a classification result that classifies the generated 3D patches into a perforation area, which is an area including the perforation, and a general area, which is an area not including the perforation, and outputting the segmentation result that segments the perforation area from the CT image based on the classification result. CT image processing method.

5. In paragraph 1, The steps for generating the above result image are: A step of generating the result image in which the perforation is indicated in the CT image by masking the CT image based on the segmentation result. A CT image processing method comprising:

6. In paragraph 1, The above perforation detection model is, Contains softmax layers and convolutional layers, The above softmax layer is, Generate a classification result that classifies a perforated area including the perforated area and a general area not including the perforated area from the generated 3D patches, and The above convolution layer, Outputting a segmentation result that segments the perforated paper area based on the classification result. CT image processing method.

7. A computer program stored in a computer-readable recording medium to execute the method of claim 1 by being combined with hardware.

8. In a CT image processing device for detecting and segmenting perforations in CT (computed tomography) images, A region of interest detection unit that detects a region of interest (ROI) in which a perforation is expected to exist from the CT image using a region of interest detection model; A three-dimensional patch generation unit for generating a plurality of three-dimensional patches including at least a portion of the detected region of interest; and A result image generation unit that inputs the plurality of 3D patches into the perforation detection model and generates a result image by extracting the perforation from the CT image based on the segmentation result from the perforation detection model. A CT image processing device including:

9. In paragraph 8, The above region of interest detection model and the above perforation detection model are, Contains a neural network model, The above region of interest detection unit, Detect the region of interest for each frame of the CT image, Determining a three-dimensional region of interest in the shooting area of ​​the CT image based on the region of interest detected for each frame above. CT image processing device.

10. In paragraph 9, The above 3D patch generation unit is, Generating the three-dimensional patches by dividing the three-dimensional region of interest into three-dimensional sub-regions smaller than the three-dimensional region of interest. CT image processing device.

11. In paragraph 8, The above perforation detection model is, Generating a classification result that classifies the generated 3D patches into a perforation area, which is an area including the perforation, and a general area, which is an area not including the perforation, and outputting the segmentation result that segments the perforation area from the CT image based on the classification result. CT image processing device.

12. In paragraph 8, The above result image generation unit is, By masking the CT image based on the segmentation result, the result image is generated in which the perforation site is indicated in the CT image. CT image processing device.

13. In paragraph 8, The above perforation detection model is, Contains softmax layers and convolutional layers, The above softmax layer is, Generate a classification result that classifies a perforated area including the perforated area and a general area not including the perforated area from the generated 3D patches, and The above convolution layer, Outputting the result of segmenting the perforation area based on the classification result. CT image processing device.

Images